User loginNavigation |
OoO JavaWe typically hear about optimistic transactional memory, so it's refreshing to see analysis-driven work on the pessimistic side in OoOJava: Software Out-of-Order Execution. Of note, they guarantee determinism:
(James C. Jenista, Yong hun Eom, and Brian Demsky) In discussions with the authors, it seems that the evaluation benchmarks were on embarrassingly parallel programs with no library dependencies, so there are many questions left to answer about the pragmatics involved. Perhaps another interesting paper in this vein is Deterministic Parallel Java which takes a type specification based approach. A question arises of software engineering: when writing a maintainable parallel programs with potential algorithmic interference, which invariants should we specify? If just basic task boundaries, OoOJ is almost like program analysis (instead of type inference) for DPJ; if non-interference guarantees, we start to get a very different picture. By lmeyerov at 2010-09-18 18:00 | LtU Forum | previous forum topic | next forum topic | other blogs | 5872 reads
|
Browse archives
Active forum topics |
Recent comments
36 weeks 3 days ago
36 weeks 3 days ago
36 weeks 3 days ago
1 year 6 weeks ago
1 year 10 weeks ago
1 year 12 weeks ago
1 year 12 weeks ago
1 year 15 weeks ago
1 year 19 weeks ago
1 year 19 weeks ago